Substrate multiplexed protein engineering facilitates promiscuous biocatalytic synthesis

Enzymes with high activity are readily produced through protein engineering, but intentionally and efficiently engineering enzymes for an expanded substrate scope is a contemporary challenge. One approach to address this challenge is Substrate Multiplexed Screening (SUMS), where enzyme activity is measured on competing substrates. SUMS has long been used to rigorously quantitate native enzyme specificity, primarily for in vivo settings. SUMS has more recently found sporadic use as a protein engineering approach but has not been widely adopted by the field, despite its potential utility. Here, we develop principles of how to design and interpret SUMS assays to guide protein engineering. This rich information enables improving activity with multiple substrates simultaneously, identifies enzyme variants with altered scope, and indicates potential mutational hot-spots as sites for further engineering. These advances leverage common laboratory equipment and represent a highly accessible and customizable method for enzyme engineering.


Supplementary
. Active-site model of RgnTDC (built from PDB ID: 4OBV). 1 Residues highlighted in purple are sites at which mutations were found that significantly altered promiscuity or improved activity. Mutations at residues highlighted in gray were found to not significantly alter promiscuity (H120, L126) or resulted in catalytically feeble enzymes (L336, T356).

Supplementary Figure 6. Retention of function (ROF) curve from SUMS of the RgnTDC
F98 site-saturation mutagenesis library. Colored bars represent relative amounts of each product formed, and diamonds represent mM total product produced, as determined by singleion retention standard curves. The wild-type sequence is denoted by a grey diamond. Relative product amounts and mM total product were averaged from all wells with the given sequence. Colored bars represent relative amounts of each product formed, and diamonds represent total product formed, based on ion count integrations of single-ion retention (SIR) channels. Parent (2B9) controls are denoted by grey diamonds, while negative controls are shown as red diamonds. Substrate panel: 5 mM each of 2-Me-indole, 4-CN-indole, 5-OMe-indole, 6-OH-indole, and indoline; and 2.5 mM 7-Clindole. The signal-to noise ratio throughout plate A was generally poor, but variant H275R was identified in this screen and subsequently validated in both substrate-multiplexed and singlesubstrate reactions. Figure 24. SUMS results for plate B of the PfTrpB globally random mutagenesis library, with 2B9 as the parent enzyme. Colored bars represent relative amounts of each product formed, and diamonds represent total product formed, based on ion count integrations of single-ion retention (SIR) channels. Parent (2B9) controls are denoted by grey diamonds, while negative controls are shown as red diamonds. Substrate panel: 5 mM each of 2-Me-indole, 4-CN-indole, 5-OMe-indole, 6-OH-indole, and indoline; and 2.5 mM 7-Clindole. Greyed-out section indicates wells where the quality of signal-to-noise was too low and the relative product distributions are dominated by noise and therefore no longer indicative of enzyme promiscuity. Figure 25. SUMS results for plate C of the PfTrpB globally random mutagenesis library, with 2B9 as the parent enzyme. Colored bars represent relative amounts of each product formed, and diamonds represent total product formed, based on ion count integrations of single-ion retention (SIR) channels. Parent (2B9) controls are denoted by grey diamonds, while negative controls are shown as red diamonds. Substrate panel: 5 mM each of 2-Me-indole, 4-CN-indole, 5-OMe-indole, 6-OH-indole, and indoline; and 2.5 mM 7-Clindole. Greyed-out section indicates wells where the quality of signal-to-noise was too low and the relative product distributions are dominated by noise and therefore no longer indicative of enzyme promiscuity. Figure 26. PfTrpB I102T single substrate activity. Activity of I102T in single substrate reactions is shown relative to 2B9 (black dashed line). Reactions consisted of 10 mM indole substrate, 30 mM Ser, and 2 µM PfTrpB in 100 mM potassium phosphate buffer (pH = 8.0) and 10% DMSO. Reactions were run in duplicate at 37 °C. Figure 27. SUMS results for the PfTrpB H275X site-saturation mutagenesis library. Colored bars represent relative amounts of each product formed, and diamonds represent total product formed, based on ion count integrations of single-ion retention (SIR) channels. The parent (2B9) sequence is denoted by a grey diamond. Substrate panel: 5 mM each of 2-Me-indole, 4-CN-indole, 5-OMe-indole, 6-OH-indole, and indoline, and 2.5 mM 7-Cl-indole. No 4-CN-Trp product was detected. No lysine, asparagine, or tyrosine mutations were sequenced from this library plate. The sampling rate of each mutation is detailed in Supplementary Table 4.

Introduction to substrate competition kinetics behind SUMS
We provide here a brief discussion on the underlying kinetics for multiplexed reactions as an initial guide for understanding substrate competition kinetics in the context of SUMS for protein engineering, particularly when increased activity on single substrates is desired. We note that Stanišić et al. 2019 provide an excellent discussion on substrate competition kinetics. 5 We also found Chou and Talalay 1977, Cornish-Bowden 1984, and Andrews 2016 to be particularly useful references. [6][7][8] While detailed understanding of substrate competition kinetics is not required to implement SUMS, we suggest that consideration of the impact of different kinetic effects on screening results will aid attempts to engineer synthetically useful enzymes. In this simplified model, KM can be approximated as a function of the rate constants involved in the formation/breakdown of the ES complex:

Eq. 3
In the case of single substrate reactions where [S] >> KM, the term kcat * [ET] = Vmax becomes the main determinant of the rate of product formation. We show below that this familiar kinetic phenomenon no longer holds for reactions where multiple substrates are present.
We found it instructive to consider substrate competition through the lens of competitive inhibition. A competitive inhibitor decreases overall enzymatic activity by binding in the enzyme active site, decreasing the availability of enzyme active sites that can catalyze the given transformation on substrate S (Eq. 4).

Eq. 2
The impact of inhibitor I with an inhibition constant KI on the rate of an enzymatic reaction can be calculated as: Eq. 5 where: .
Eq. 6 In a multiplexed reaction, substrates will mutually inhibit each another by competing for active site binding in a fashion analogous to competitive inhibition. Given the simple model below of SA and SX competing to form products PA and PX (Eq. 7), respectively, the additional rate constant k4 for the formation of PX is the only distinction between this model and the previous competitive inhibition model (Eq. 4).
SX becomes a classical competitive inhibitor when k4 = 0, where KMX = KI. This means KMX can be defined as the Michaelis-Menten constant for a given SX and describes the inhibition of SX on SA, as shown in Eq. 8.

Eq. 8
Therefore, for any given number of additional substrates, the initial velocity of formation of PA can be described as:

Supporting data for substrate competition kinetic model
Many synthetic applications of enzymes use single-substrate reaction conditions. In the context of single-substrate reactions with high substrate concentrations, kcat is an excellent predictor of relative activity, as measured by observed total turnover numbers for different chlorinated tryptophan analogs ( Supplementary Fig. 38a, data from Ref 2). Catalytic efficiency, kcat/KM, is an ineffective predictor of single-substrate activity ( Supplementary Fig. 38b). Conversely, in a reaction where substrates compete for an enzyme active site, the ratio of the substrates' kcat/KM values can be used to predict relative product abundance. From an example three-substrate multiplexed reaction of RgnTDC, we show that the relative ratios of the three products do not correlate well to the substrates' relative kcat's (Supplementary Fig. 39a) but correlate very well with substrate kcat/KM's, as determined from single-substrate kinetic measurements ( Supplementary Fig. 39b). Notably, the KM values for Trp, 4-Br-Trp, and 6-Cl-Trp with RgnTDC are 1.6 mM, 3.4 mM, and 0.14 mM, respectively. The reaction depicted in Supplementary Fig. 39b was conducted with 5 mM of each Trp analog, demonstrating that product profiles reflect the ratio of each substrate's catalytic efficiency, even at concentrations that exceed a substrate's KM. SUMS methods are compatible with high substrate concentrations, which are typically desirable for synthetic applications.

Repression of inhibition
During library screening, some variants may acquire mutations that increase the abundance of one or more products. Several distinct kinetic scenarios can provide this outcome. In the simplest case, the mutations may increase the kcat with a substrate, leading to more product formed. Alternatively, mutations may leave kcat minimally changed but selectively decrease the KM with a substrate, making it a more effective competitive inhibitor in a multiplexed setting and leading to increased occupancy of the available active sites. Both kinetic changes provide a classically activated enzyme, where mutation increases the catalytic efficiency, kcat/KM. However, there is an alternative scenario that may arise in substrate multiplexed settings where the relative abundance of a product increases, but there is either no change or a decrease in the catalytic efficiency of the enzyme. This phenomenon arises when two (or more) substrates in competition react with significantly different catalytic efficiencies. As has been described above, the substrate with the higher catalytic efficiency will dominate the enzyme active site. If a mutation decreases the catalytic efficiency for both substrates but is much more deleterious with the originally preferred substrate, then the occupancy of the active site in competition will shift relative to the parent enzyme to favor the less-reactive substrate. In effect, the ability of a 'good' substrate to inhibit reaction with 'bad' substrates is diminished. Consequently, the relative amount of product formed from a poorly active substrate can increase. We describe this phenomenon as activation through "repression of inhibition." Notably, the apparent increases in relative activity caused by repression of inhibition detected by a multiplexed screen do not translate to single substrate conditions, since there has been no improvement in activity with the poor substrate. This phenotype is reproducible, and not an indication of errors or false positives in the screening process.
To help illustrate this subtle concept, we have generated a hypothetical scenario in Supplementary Fig. 40. The hypothetical Parent enzyme strongly favors formation of product B in a multiplexed setting. For Variant 1, the KM for substrate B has increased 100-fold, suppressing its ability to inhibit substrate A. Consequently, Variant 1 produces far more product A than the Parent in the multiplexed assay, as determined by the relative kcat/KM values for each substrate-enzyme pair. However, parent and variant 1 would form product A at the same rate in a single-substrate reaction, since both enzymes have the same kcat and KM for substrate A.
In the more complex case of Variant 2, the KM for substrate B has again increased 100-fold, suppressing its ability to inhibit substrate A. Along with this change, we considered a decrease in kcat and increase in KM for substrate A. In the multiplexed assay, Variant 2 will generate more product from substrate A. However, single-substrate reactions will reveal that Variant 2 actually has decreased catalytic efficiency with substrate A relative to Parent. At first glance, single-substrate and multiplexed results may appear to contradict one another, but we contend through the examples shown here that the repression of inhibition phenotype can be rationalized using standard kinetic analysis.
We have not yet determined a simple means for predicting whether improvements in activity identified by SUMS are due to an increase in catalytic efficiency or merely repression of inhibition. However, we observe this phenotype most frequently when screening on substrates that have significant differences in their relative reactivity. During the assay design phase of SUMS (detailed below), one can choose substrates with similar reactivity to limit the impact of the repression of inhibition phenomenon. Ultimately, validation of potential hits is a core step in protein engineering, regardless of whether single-substrate screening or SUMS was used. We assert that variants that demonstrate a shift in product profile (such as the TrpB variant H275R, which showed a repression of inhibition phenotype) can point to mutations that influence substrate specificity and are therefore interesting candidates for further investigation.

Guide to optimization of SUMS method
We provide here a discussion on possible optimization routes for fine-tuning a substrate multiplexed screen. We acknowledge there are additional options for optimization but will highlight ones we felt had the greatest impact on our development of SUMS for protein engineering.

Choice of substrates
While there is no "right" or "wrong" panel of substrates, the substrates chosen for SUMS will directly impact the information gained from screening. This list is not intended to be comprehensive, and any substrate panels should be tested with parent enzyme to confirm that all desired products can be resolved as anticipated.

a. Number of substrates
In theory, any number of substrates can be added to a multiplexed substrate mixture. We chose <10 substrates for the reported screens for simplicity and to ease interpretation of the data. Additional substrates may be considered according to the time constraints of the researcher, the degree of mutual substrate inhibition with the selected substrates, and any potential instrumental limitations.

b. Substrate/product masses
The implementation of SUMS described here leveraged MS-based peak integration to quantitate distinct products. Although chromatographic resolution of products of the same mass could work in a SUMS setting ( Supplementary  Fig. 41), we chose substrates that will yield products with unique m/z's.

c. Relative substrate activity
Since highly active substrates can inhibit the formation of products from less active substrates, the addition of both highly active and inactive substrates to a substrate mixture may decrease the amount of information obtained from a screen. In principle, a SUMS setup will have maximum sensitivity when each of the substrates is a relatively poor substrate for the native enzyme, such that none is a good competitive inhibitor and there is a large potential dynamic range of improved activities. However, differences in baseline activity do not prevent successful application of SUMS. Substrates with which the parent enzyme has no detectable activity can be added to the screening mixture to monitor for variants that gain new activity. The presence of more reactive substrates may obscure low levels of activity that arise, but modulation of relative substrate concentrations can account for differences in activities. For both RgnTDC and TrpB engineering, the concentration of the most highly active substrate(s) was reduced relative to the concentration of other substrates.

Screening time
For a single-substrate screen to be sensitive to improvements in activity, the parent enzyme must not reach full conversion of substrate to product. However, in a multiplexed screen, the reaction rate for different substrates may vary greatly. It is possible that some substrates may reach high or even full conversion while others show only trace activity. One may thus choose conditions under which all desired products are observed, possibly resulting in full consumption of certain substrates. As reactions depart the initial velocity regime, the relative product ratios will shift away from the ratios of substrates' kcat/KM values, as shown in Supplementary Fig. 42a. We therefore describe the distribution of products as the SUMS promiscuity profile rather than a specificity profile, since relative rates of product formation are not necessarily assessed. In this way, the screen can identify variants that both retain existing activity and improve activity on poor substrates.

a. Unimolecular reactions
To help offset the impact of mutual substrate inhibition, the relative concentrations of the various substrates can be adjusted. As shown in Supplementary Fig. 42b, the inhibitory capacity of highly reactive substrates can be reduced by decreasing their concentration. Such an approach increases the abundance of poorer products ( Supplementary Fig. 42b, ~10fold increase in total product formed for 4-Br-Trp and 2-Me-Trp) while maintaining an activity threshold for the substrates with high baseline activity.

b. Higher order reactions
For enzymes that catalyze bimolecular and higher order reactions, early timepoint product distributions can be 'captured' by limiting the stoichiometry of a non-multiplexed substrate Knorrscheidt et al. 2021 describe such an approach with limiting concentrations of H2O2 for their peroxygenase reactions. 9 We chose to use a limiting amount of L-serine (Ser) in the bimolecular reaction catalyzed by TrpB to ensure that tryptophan production stalled with respect to indole consumption. In this way, optimization of reaction time and catalyst loading was simplified. Alternatively, an excess of Ser could have been used to increase assay sensitivity to low abundance products.

Hit Identification
Unlike in single-substrate screens, where identifying the variant with the highest activity is the singular goal, in SUMS there may be multiple classifications of a 'hit' or a variant that merits further investigation. Here, we will discuss briefly how we classified hits in our engineering campaigns of RgnTDC and TrpB-2B9.
For RgnTDC engineering, we looked primarily for two types hits from screening: 1) variants with boosts in total activity and 2) variants with a boost in activity on at least one individual substrate in the screen. As we did not find any generally activating mutations, the hits we investigated fell primarily under the second class of hits. Variants such as W349K and L355A showed enhanced activity with previously low-activity Trp analogs, and thus were further investigated.
We took a different approach to TrpB engineering, as we were no longer looking in the active site, where we could expect to easily observe drastic changes in activity and specificity. Therefore, along with the two previously described hit classifications, we looked for an additional type of hit: 3) variants with shifts in the product distribution (regardless of whether activity for any substrate(s) increases or decreases). We surmised that such shifts would be indicative of mutations that are involved with substrate discrimination and therefore are likely influencing the enzyme active site. During screening, we found the generally activated I102T, which falls under the first hit classification, and H275R, which decreased total activity but shifted the product profile, falling under the third hit classification.

General experimental methods
Chemicals and reagents were purchased from commercial suppliers (Sigma-Aldrich, VWR, Chem-Impex International, Alfa Aesar) and used without further purification. E. coli (BL21 (DE3)) cells were used for protein expression. An Eppendorf E-porator (2500 V) was used for transformations. New Brunswick I26R, 120 V/60 Hz shakers (Eppendorf) were used for cell growth. Cell disruption via sonication was performed with a Sonic Dismembrator 550 (Fisher Scientific) sonicator. Optical density measurements were collected on a UV-2600 Shimadzu spectrophotometer (Shimadzu). UPLC-MS data were collected on an Acquity UHPLC with an Acquity QDA MS detector (Waters). Column separations were performed on an Isolera One Flash Purification system (Biotage). NMR data were collected on a Bruker 500 MHz spectrometer. High resolution mass data were collected with a Q Extractive Plus Orbitrap (NIH 1S10OD020022-1) instrument with the samples ionized by ESI. Kinetic data were fit with PRISM 8 Graphpad software.

Cloning and expression of RgnTDC site-saturation libraries
A codon-optimized copy of the Ruminococcus gnavus tryptophan decarboxylase (RgnTDC) gene with a C-terminal His-tag was previously purchased 2 as a gBlock from Integrated DNA Technologies. This DNA fragment was inserted into a pET22b vector via Gibson assembly. 10 BL21 E. coli cells were subsequently transformed with the resulting cyclized DNA product via electroporation. After 30 min of recovery in Luria-Burtani (LB) media at 37 °C, cells were plated onto LB plates with 100 µg/mL ampicillin (AMP) and incubated overnight. Single colonies were used to inoculate 5 mL TB + 100 µg/mL AMP and grown overnight at 37 °C. pET22b-RgnTDC plasmid was purified via the Zymo Research Plasmid Prep kit and eluted in 20 µL sterile Milli-Q H2O. Plasmids were stored at -20 °C.
Based on a previously reported model of an RgnTDC-Trp complex (PDB ID: 4OBV), 1,2 active site residues in the RgnTDC active site were chosen for mutagenesis. The sites were: Phe98, Val99, His120, Leu126, Leu336, Leu339, Trp349, Leu355, and Thr356 ( Supplementary Fig. 5). Primers were purchased from Integrated DNA Technologies. For each site of mutation, three primers encoding the degenerate codons NDT, VHG, and TGG at the codon of interest were mixed in a 12:9:1 ratio, respectively. 4 Each gene library was amplified first as two separate fragments and then combined via polymerase chain assembly (PCA) 11 to form full-length RgnTDC gene mutagenized at the site of interest. The corresponding genes were then inserted into a pET22b vector as described above and then transformed into BL21(DE3) E. coli cells and plated on LB + 100 µg/mL AMP agar plates.
Individual colonies were picked and used to inoculate wells containing 600 µl TB + 100 µg/mL AMP in a 96-well plate. One column of the plate was used as an internal control with five wild-type colonies, one sterile control, and two non-RgnTDC pET22b negative controls. Plates were grown at 37 °C for 16 h at 200 RPMs. 20 µl of this starter culture plate was used to inoculate 630 µl TB + 100 µg/mL AMP. These expression plates were grown at 37 °C for 3 h until OD600 ~ 1.0. Plates were then placed on ice for 1 h. Expression was induced by addition of 50 µl TB + 100 µg/mL AMP + 14 mM IPTG + 7 mM indole. Plates were then grown at 23 °C for 16 h at 200 RPM and subsequently spun down at 4000 xg for 15 min and the supernatant discarded. Plates containing cell pellets were stored at -20 °C.

Cloning of PfTrpB globally random mutagenesis libraries
The plasmid encoding an engineered β-subunit of tryptophan synthase from Pyrococcus furiosus, PfTrpB 2B9 (2B9), was obtained from the Arnold lab. 12 The mutations in this variant were previously shown to confer kinetic and spectroscopic properties that mimic native allosteric activation by the α-subunit of tryptophan synthase. 3 The 2B9encoding plasmid was used as a template for global random mutagenesis to prepare error-prone libraries. PCR amplification of the 2B9 gene was conducted with the addition of MnCl2 (100 or 150 µM) to induce mutagenesis. The amplified products were purified by gel electrophoresis and extracted from the gel using a Zymoclean Gel DNA Recovery Kit. The resulting gene libraries were inserted into a pET22(b)+ plasmid via Gibson assembly 10 and transformed into E. coli BL21(DE3) cells via electroporation. Cells were recovered in TB media for 30 minutes at 37 °C and 220 RPM. Cells were plated onto LB + 100 µg/mL AMP agar plates and incubated overnight at 37 °C.

Cloning of PfTrpB H275X site-saturation library
To construct a site-saturation mutagenesis library at the H275 site, three primers with the degenerate codons NDT, VHG, and TGG were purchased from IDT and mixed in a 12:9:1 ratio. 4 The 2B9 gene was amplified with this primer mixture, resulting in two gene fragments, which were combined via PCA to generate a full-length gene. 11 The gene was purified by gel electrophoresis and extracted using a Zymoclean Gel DNA Recovery Kit. The resulting gene library was inserted into a pET22(b)+ plasmid via Gibson assembly and transformed into E. coli BL21(DE3) cells via electroporation. Cells were recovered in TB media for 30 minutes at 37 °C and 220 RPM. Cells were plated onto LB + 100 µg/mL AMP agar plates and incubated at 37 °C overnight.

Screening of PfTrpB libraries
Individual colonies were picked and used to inoculate wells containing 600 µl TB + 100 µg/mL AMP in a 96-well plate. One column of the plate was used as an internal control with five parent (2B9) colonies, one sterile control, and two non-PfTrpB pET22b negative controls. Plates were grown at 37 °C for 16 h at 200 RPM. Expression cultures in a fresh 96-deep well plate containing TB + 100 µg/mL AMP (630 µL) were inoculated using starter culture (20 µL) and grown at 37 °C and 180 RPM until cultures reached an OD600 of at least 0.6 (approximately 3 hours). Expression cultures were chilled on ice for 30 minutes then inoculated with IPTG (1 mM final concentration). Expression plates were incubated at 23 °C for 16 h at 180 RPM then centrifuged at 4000 xg for 15 min and the supernatant discarded. Cell pellets were stored at -20 °C until lysed.
A 96-well plate was loaded with 20 µL substrate mixture (final concentration of 5 mM each 2-methylindole, 4-cyanoindole, 5-methoxyindole, 6-hydroxyindole, and indoline, plus 2.5 mM 7-chloroindole). All indole stocks were prepared in DMSO. For globally random mutagenesis library plate A, potassium phosphate buffer (50 mM, pH = 8.0, 160 µL) containing L-serine (5 mM final concentration) was added, followed by heat-treated lysate (20 µL). For subsequent plates, lysate volume was increased to 50 µL and buffer volume reduced to 130 µL. Reactions were set up such that the DMSO cosolvent comprised 10% of the final reaction volume (200 µL). Reactions were run at room temperature (25 °C) for 2.5 h and were quenched with 200 µL acetonitrile containing 0.1 M HCl and 1 mM tryptamine (as internal standard). Plates were spun down at 4000 xg at 4°C for 20 min. A 200 µL aliquot of each quenched reaction was filtered into a 96-well plate for analysis by UPLC-MS. Product formation was quantified by integration of peaks on single ion retention (SIR) channels corresponding to each expected product, normalized against the tryptamine internal standard.

Expression of RgnTDC and PfTrpB variants
Cells from wells containing putatively activated RgnTDC or PfTrpB variants were inoculated into 5 mL TB + AMP and grown for 16 h at 37 °C. Each variant strain culture was used to inoculate 500 mL TB + AMP, which was subsequently grown at 37 °C for 3 h, or until OD600 ~ 1.5. Cultures were chilled on ice for 30 min, and then induced with 1 mM IPTG. Cells expressing RgnTDC were additionally supplemented with 0.5 mM indole, as this was found to boost cell growth. Cultures expressed for 16 h at 23 °C. The next morning, cells were pelleted via centrifugation at 4000 xg for 15 min. Supernatants were bleached and discarded and pellets were weighed and stored at -20 °C until lysis.
To purify enzyme variants, cell pellets were thawed at room temperature and then resuspended in lysis buffer: 50 mM potassium phosphate buffer (pH = 8.0), 1 mg/mL Hen Egg White Lysozyme (GoldBio), 0.2 mg/mL DNaseI (GoldBio), 1 mM MgCl2, and 300 μM pyridoxal 5′-phosphate (PLP). A volume of 4 mL of lysis buffer per gram of wet cell pellet was used. After 30 min of shaking at 37 °C, RgnTDC lysis suspensions were disrupted using sonication (5 min; 0.8 s on, 0.2 s off at a power setting of 5). PfTrpB lysis suspensions were heat treated at 75 °C for 15 min as an alternative to sonication. The resulting lysate was spun down at 75,000 xg to pellet cell debris. Ni/NTA beads (GoldBio) were added to the supernatant and incubated on ice for 30 to 60 min prior to purification by Ni-affinity chromatography. The column was washed with 3 column volumes of wash buffer A (20 mM imidazole, 50 mM potassium phosphate buffer (pH = 8.0)) and with 2 column volumes of wash buffer B (40 mM imidazole, 50 mM potassium phosphate buffer (pH = 8.0); PfTrpB variants only), and the proteins were eluted with an elution buffer (250 mM imidazole, 50 mM potassium phosphate buffer (pH = 8.0)). Elution of the desired protein product was monitored by the disappearance of its bright yellow color (resulting from the PLP cofactor) from the column. The protein product was dialyzed to < 1 μM imidazole, dripped into liquid nitrogen to flash freeze, and stored at -80 °C. The concentration of protein was determined by Bradford assay upon thawing. Generally, this procedure yielded 50 -200 mg per L culture for RgnTDC variants and 200 -500 mg per L culture for PfTrpB variants.

Turnover analysis of RgnTDC variants
RgnTDC variants were thawed on ice from storage at -80 °C and then centrifuged at 15,000 xg for 5 min to pellet aggregated protein. The supernatant was then diluted between 1:1 and 1:100 in 50 mM potassium phosphate buffer (pH = 8.0) (depending on enzyme concentration). Reactions were run in plastic 96-well plates with 10 mM Trp analog, either 33 nM or 1 μM TDC variant, and 1000 equivalents PLP relative to TDC with a total volume of 100 μL in 50 mM potassium phosphate buffer (pH = 8.0). Reactions were incubated for 16 h at 37 °C. Reactions were quenched with 200 μL acetonitrile and further diluted with 100 μL H2O. Quenched reactions were centrifuged at 15,000 xg for 10 min prior to analysis by UPLC-MS. Enzymatic activity was quantified by integrating the substrate and product UV absorbance peaks at 280 nm. Reactions were run in triplicate, with replicate reactions run on different days with freshly thawed enzyme (Fig 2g).

[PLP] vs. activity for RgnTDC variants
RgnTDC variants were thawed and prepared as described above. 10 mM 6chlorotryptophan was added to a solution with varying molar equivalents of PLP relative to TDC (500 nM, 5 μM, 50 μM, or 500 μM). TDC was added such that the final concentration was 50 nM and the solution was diluted up to 100 μl with 50 mM potassium phosphate buffer (pH = 8.0). Reactions were incubated for 16 h at 37 °C. Reactions were quenched with 200 μL acetonitrile and diluted with 100 μL H2O. Quenched reactions were centrifuged at 15,000 xg for 10 min prior to UPLC-MS injection of the supernatant. Enzymatic activity was quantified by integrating the substrate and product UV absorbance peaks at 280 nm. Reactions were run in duplicate, with replicate reactions run on separate days with freshly thawed enzyme ( Supplementary Fig. 22).

RgnTDC substrate multiplexed timecourse reaction
RgnTDC variants were thawed and prepared as described above. Reaction conditions included 0.25 μM RgnTDC, 2.5 μM PLP, and one of two sets of substrate mixes: 2.5 mM tryptophan, 2-methyltryptophan, 4-bromotryptophan, and 6-chlorotryptophan; or 0.25 mM tryptophan and 6-chlorotryptophan, 2.5 mM 2-methyltryptophan and 4-bromo-tryptophan. Reactions were filled to 100 μl with 50 mM potassium phosphate pH = 8.0. Reactions were conducted in triplicate and incubated for 4 h at 37 °C. Reactions timepoints were taken by quenching 10 μl of the reaction solution in 190 μl of acetonitrile, followed by dilution with 200 μl H2O. Quenched reactions were centrifuged at 15,000 xg for 10 min prior to UPLC-MS injection of the supernatant. Product abundance was quantified via single-ion retention and fit to product standard curves ( Supplementary Fig. 16).

Michaelis-Menten analysis of PfTrpB variants
PfTrpB variants were thawed on ice from storage at -80 °C and then centrifuged at 15,000 xg for 5 min to pellet aggregated protein. The supernatant was diluted in 100 -200 mM potassium phosphate buffer (pH = 8.0) to obtain sufficiently dilute concentrations for initial velocity kinetic measurements.
Indole: Rate of formation of Trp from indole and Ser was measured at 290 nm, as described previously. 14 PfTrpB variants (0.34 -1.57 μM final concentration) in 200 mM potassium phosphate buffer (pH = 8.0) were pre-incubated in quartz cuvettes at 37 °C. Indole (3 -400 μM final concentration) was added, followed by Ser (2.5 mM for 2B9 and H275E, 25 mM for H275R). Indole stocks were prepared using DMSO, such that the final reaction volume (400 μL) was comprised of 5% DMSO. Cuvettes were inverted to mix, and absorbance at 290 nm was collected for 2 min. All replicate reactions were run on two different days with fresh enzyme. Initial velocity slopes were calculated and fit using the revised Michaelis-Menten equation as described by Johnson 2019. 13 Indoline: Formation of 2,3-dihydroiisotryptophan (DIT) from indoline and Ser was measured by integration of UPLC peaks at 254 nm. A calibration curve was used to convert integration of the UPLC peak to a concentration of DIT ( Supplementary Fig. 37). Reaction conditions included 1 -50 mM indoline, 10 mM Ser, 1 -2 µM PfTrpB, and 5% DMSO in 100 mM potassium phosphate buffer (pH = 8.0) in 1.7 mL epi-tubes. The enzyme and substrate solutions were equilibrated to 37 °C prior to mixing. Reaction aliquots (50 μL) were quenched in acetonitrile (125 μL) at various time points (5 -60 min) and diluted with 75 μL water (1:5 dilution factor). Quenched samples were centrifuged (15,000 xg, 5 min) and analyzed by UPLC-MS. All replicate reactions were run on three different days with fresh enzyme. Initial velocity slopes were calculated and fit using the revised Michaelis-Menten equation as described by Johnson 2019. 13

UV-Vis spectroscopy of PfTrpB variants
PfTrpB variants were thawed on ice from storage at -80 °C and then centrifuged at 15,000 xg for 5 min to pellet aggregated protein. The supernatant was diluted in 200 mM potassium phosphate buffer (pH = 8.0) to obtain a final concentration of 300 µM. A baseline spectrum (600-250 nm, 1 nm interval, fast scan) was collected on 200 mM potassium phosphate buffer (284 µL, pH = 8.0) at 25 °C. PfTrpB (20 µL) was added to the cuvette for a final concentration of 20 µM enzyme and an enzyme-only spectrum was collected. To study Ser binding and rate of the shunt pathway, Ser (12.5 µL, 20 mM final concentration) was added, and the cuvette was inverted to mix. Spectra were collected every 10 minutes for 12 scans (110 minutes total) to track pyruvate formation (320 nm) over time. To study Trp binding, Trp was titrated into a sample of 20 µM PfTrpB and a spectrum was collected at each concentration (0-1200 µM).

Protein crystallography of PfTrpB-H275E
Crystals of the H275E PfTrpB 2B9 variant were grown in sitting drops against a 1-mL reservoir containing 13-21% PEG3350 and 0.1 M Na HEPES buffer (pH = 7.85), with mother liquor containing 2 µL of H275E (8.0 or 15 mg/mL) and 2 µL of well solution. Crystals grew over the course of several days and were stable in the dark over several weeks. Trp-and 4-Cl-Trp-bound crystals were prepared by addition of solid Trp or 4-Cl-Trp (approx. 1 mg) to pre-formed crystals. Crystals were cryoprotected by dredging through Fomblin-Y then through Paratone-N and flash frozen in liquid N2. Data were collected remotely at the Argonne National Laboratory Advanced Photon Source on either beamline 21ID-D (PDB ID: 7RNQ) or beamline 23ID-B (PDB ID: 7RNP, 7ROF). Data were integrated and scaled using XDS and AIMLESS. 15,16 Structures were solved by molecular replacement with PHASER, using CCP4. 17,18 Search models for each structure were comprised of a single monomer of a previously solved PfTrpB 2B9 structure, where the H275 site was deleted, ligands and waters were removed, and the remaining model was subjected to geometric refinement in Refmac5. PDB ID: 6AM7 was used as the search model for the unbound H275E structure, and PDB ID: 6AM8 was used as the search model for the Trp-and 4-Cl-Trp-bound structures. Model building was performed in Coot and refinement was conducted using Refmac5. The TLS motion server was used to calculate TLS operators; we selected 10 operators for each chain in the asymmetric unit. 19 The MolProbity server was used to evaluate the structure prior to final refinement. 19,20 Crystallographic and refinement statistics are reported in Supplementary  Table 3.

Tryptophan analog biosynthesis and purification
For each tryptophan analog, the corresponding indole derivative (~1 mmol) was added to a 100 mL pressure flask and dissolved in MeOH (5 mL). L-Serine (3 mmol) was added, and the resulting solution was diluted to just under 100 mL with 50 mM potassium phosphate buffer (pH = 8.0). PLP was added such that the final concentration was 300 μM. Then, 2B9 was added at 0.1% mol catalyst relative to the indole analog. The solution was incubated at 75 °C for 16 h. Following UPLC-MS analysis of conversion, the solution was heat-treated at 90 °C for 30 min. Solutions were evaporated down to 10 mL, filtered, and run over C18 via flash chromatography with H2O/MeOH. Tryptophan products typically eluted between 10-35% MeOH. Product-containing fractions were combined and evaporated down to ~2-3 mL, where the solutions were transferred to pre-tared flasks, evaporated to dryness, resuspended in H2O, flash-frozen, and lyophilized for 24 -48 h. Resulting powders were weighed and submitted for 1 H-NMR analysis, with final reported yields taking hydration states into account (all tryptophan analog products were isolated as hydrates).

5-acetyltryptophan
7-iodotryptophan was isolated as a white powder (94.3 mg) in 79% yield. 1 H NMR (500 MHz, MeOD) δ 7.75 (dd, J = 7.9, 0.9 Hz, 1H), 7.54 (d, J = 7.4 Hz, 1H), 7.30 (s, 1H), 6.88 (t, J = 7.7 Hz, 1H), 3.86 (dd, J = 9.1, 4.2 Hz, 1H), 3.50 (dd, 1H), 3.16 (dd, 1H). Spectrum matched previously reported spectra. 21 Cascade synthesis and isolation of tryptamines 4-6 mmol (1.4 mmol for 6-chloroindole) of the corresponding indole analog was added to a 1 L Erlenmeyer flask and dissolved in 20 mL MeOH. 12 mmol serine was added, and the resulting solution was diluted up to just under 500 mL with 50 mM potassium phosphate buffer pH = 8.0. PLP was added such that the final concentration was 300 μM. Then, H275E was added at 0.05% mol catalyst relative to the indole analog (0.2 mg/mL). The solution was incubated at 75 °C for 16 h. (H275E was found to be activating at 75 °C, Supplementary Fig. 33). Following UPLC-MS analysis of conversion, the solution was cooled to 37 °C, upon which RgnTDC was added at 0.02 -0.2% mol catalyst relative to the indole (0.09 -0.9 mg/mL). The solutions were incubated at 37 °C for 24 h. Solutions were then evaporated down to 50 -100 mL. To break emulsions, the solutions were acidified with 6 M HCl until pH < 1, 100 mL ethyl acetate (EtOAc) was added, and the resulting mixtures were centrifuged at 4000 x g for 10 min. These solutions were added to a separatory funnel, the aqueous layer was drained, and the organic layer removed. This was repeated twice more, with 2 mL 6 M HCl added in between extractions. Then, the aqueous layer was alkalized with 6 M NaOH until pH > 12.* Tryptamine products were then extracted 3x with 150 mL EtOAc, with 2 mL 6 M NaOH added in between extractions to the aqueous layer. Organic layers were pooled, dried with sodium sulfate, filtered, and evaporated down to 5-10 mL. Solutions were transferred to 20 mL scintillation vials, evaporated to near dryness (tryptamines were observed as liquids at 50 °C), and dried under vacuum overnight. Dried samples were weighed and submitted for 1 H and 13 C NMR analysis.

6-nitrotryptamine
F98V RgnTDC was used as 0.02% mol catalyst for the reaction with 6-nitroindole. Purification was carried out as described above until the solution was basified. The bright yellow solution was turned bright red upon addition of NaOH, and a yellow precipitate was observed. The precipitate was found to be insoluble in both diethyl ether and EtOAc and was filtered from both the aqueous and organic layers. Upon washing with cold water and diethyl ether, a yellow solid was obtained. The remaining aqueous solution was extracted 3x with diethyl ether, resulting in a pale-yellow organic layer. The organic extraction was evaporated dry, and the resulting yellow solid was combined with the filtered solid via resuspension in water. The resuspension was flash frozen and lyophilization was carried out to remove water. 6-nitrotryptamine was isolated as an initially brown powder (which after 2 weeks turned yellow) (495.5 mg) in 47% yield. Interestingly, this product produces a deep red color when dissolved in solution, which seems to be indicative of the amine group being deprotonated. 1 H NMR (500 MHz, ACN) δ 8.32 (d, J = 2.2 Hz, 1H), 7.81 (dd, J = 8.8, 2.1 Hz, 1H), 7.57 (d, J = 8.8 Hz, 1H), 7.52 (s, 1H), 2.90 -2.82 (m, 4H). Spectrum matched previously reported spectra. 2